Managing guardrails

Guardrail activity across deployed agents on the Agents page, showing which agents use guardrails, the actions enforced, and recent guardrail events.

The Guardrails tab provides a centralized view of guardrail activity across your deployed agents, helping operators ensure that agents behave reliably, securely, and in alignment with organizational policies. It surfaces runtime guardrail data in one place, from configuration coverage to enforcement outcomes.

The tab is organized around three stages:

• Configure guardrails — add tool-level guardrails to your agents.
• Monitor — review guardrail activity at runtime, tracking how often guardrails trigger and what actions they enforce.
• Fine-tune guardrails — use evaluation data to test, refine, and optimize guardrail behavior over time.

Agents Guardrails Usage

The pie chart shows the proportion of agents in your tenant that have guardrails configured versus those that do not, based on execution data. Only agents that have run at least once are reflected in this chart. Use this view to assess governance coverage across your deployed agent ecosystem and identify agents that may require guardrail configuration.

Top Invoked Actions

The stacked bar chart shows the volume and breakdown of guardrail actions enforced across your agents. Each bar represents a deployed agent, split by action type: Block, Escalate, and Filter. A filter lets you display the top 5 or top 10 agents by total guardrail invocations, consistent with other monitoring charts on the Agents page.

Guardrails details

The table lists all deployed agents with their guardrail activity. You can filter the table to show agents with guardrails, agents without guardrails, or all agents. A time range filter lets you scope the data to a selected period.

Column	Description
# of guardrails	The number of guardrails configured for the agent.
Most recent guardrail	The name of the guardrail most recently triggered during agent execution.
Guardrail type	The type classification of the guardrail.
Scope	The level at which the guardrail applies: agent, LLM, or tool.
Execution stage	Whether the guardrail runs before or after the agent, tool, or LLM execution: Pre-execution or Post-execution.
Most recent action	The action enforced by the most recently triggered guardrail: Escalate, Log, Filter, or Block.
Action result	The outcome of the most recently enforced action: Continue, Reject, Log, Filter, or Block.